Python Elasticsearch Client

Official low-level client for Elasticsearch. Its goal is to provide common ground for all Elasticsearch-related code in Python; because of this it tries to be opinion-free and very extendable.

For a more high level client library with more limited scope, have a look at elasticsearch-dsl - it is a more pythonic library sitting on top of elasticsearch-py.


The library is compatible with all Elasticsearch versions since 0.90.x but you have to use a matching major version:

For Elasticsearch 2.0 and later, use the major version 2 (2.x.y) of the library.

For Elasticsearch 1.0 and later, use the major version 1 (1.x.y) of the library.

For Elasticsearch 0.90.x, use a version from 0.4.x releases of the library.

The recommended way to set your requirements in your or requirements.txt is:

# Elasticsearch 2.x

# Elasticsearch 1.x

# Elasticsearch 0.90.x

The development is happening on master and 1.x branches, respectively.

Example Usage

from datetime import datetime
from elasticsearch import Elasticsearch
es = Elasticsearch()

doc = {
    'author': 'kimchy',
    'text': 'Elasticsearch: cool. bonsai cool.',
res = es.index(index="test-index", doc_type='tweet', id=1, body=doc)

res = es.get(index="test-index", doc_type='tweet', id=1)


res ="test-index", body={"query": {"match_all": {}}})
print("Got %d Hits:" % res['hits']['total'])
for hit in res['hits']['hits']:
    print("%(timestamp)s %(author)s: %(text)s" % hit["_source"])


This client was designed as very thin wrapper around Elasticseach’s REST API to allow for maximum flexibility. This means that there are no opinions in this client; it also means that some of the APIs are a little cumbersome to use from Python. We have created some Helpers to help with this issue.

Persistent Connections

elasticsearch-py uses persistent connections inside of individual connection pools (one per each configured or sniffed node). Out of the box you can choose to use http, thrift or an experimental memcached protocol to communicate with the elasticsearch nodes. See Transport classes for more information.

The transport layer will create an instance of the selected connection class per node and keep track of the health of individual nodes - if a node becomes unresponsive (throwing exceptions while connecting to it) it’s put on a timeout by the ConnectionPool class and only returned to the circulation after the timeout is over (or when no live nodes are left). By default nodes are randomized before being passed into the pool and round-robin strategy is used for load balancing.

You can customize this behavior by passing parameters to the Connection Layer API (all keyword arguments to the Elasticsearch class will be passed through). If what you want to accomplish is not supported you should be able to create a subclass of the relevant component and pass it in as a parameter to be used instead of the default implementation.


Since we use persistent connections throughout the client it means that the client doesn’t tolerate fork very well. If your application calls for multiple processes make sure you create a fresh client after call to fork.

Automatic Retries

If a connection to a node fails due to connection issues (raises ConnectionError) it is considered in faulty state. It will be placed on hold for dead_timeout seconds and the request will be retried on another node. If a connection fails multiple times in a row the timeout will get progressively larger to avoid hitting a node that’s, by all indication, down. If no live connection is availible, the connection that has the smallest timeout will be used.

By default retries are not triggered by a timeout (ConnectionTimeout), set retry_on_timeout to True to also retry on timeouts.


The client can be configured to inspect the cluster state to get a list of nodes upon startup, periodically and/or on failure. See Transport parameters for details.

Some example configurations:

from elasticsearch import Elasticsearch

# by default we don't sniff, ever
es = Elasticsearch()

# you can specify to sniff on startup to inspect the cluster and load
# balance across all nodes
es = Elasticsearch(["seed1", "seed2"], sniff_on_start=True)

# you can also sniff periodically and/or after failure:
es = Elasticsearch(["seed1", "seed2"], sniff_on_start=True, sniff_on_connection_fail=True, sniffer_timeout=60)

SSL and Authentication

You can configure the client to use SSL for connecting to your elasticsearch cluster, including certificate verification and http auth:

from elasticsearch import Elasticsearch

# you can use RFC-1738 to specify the url
es = Elasticsearch(['https://user:secret@localhost:443'])

# ... or specify common parameters as kwargs

# use certifi for CA certificates
import certifi

es = Elasticsearch(
    ['localhost', 'otherhost'],
    http_auth=('user', 'secret'),


By default SSL certificates won’t be verified, pass in verify_certs=True to make sure your certificates will get verified. The client doesn’t ship with any CA certificates; easiest way to obtain the common set is by using the certifi package (as shown above).

See class Urllib3HttpConnection for detailed description of the options.


elasticsearch-py uses the standard logging library from python to define two loggers: elasticsearch and elasticsearch.trace. elasticsearch is used by the client to log standard activity, depending on the log level. elasticsearch.trace can be used to log requests to the server in the form of curl commands using pretty-printed json that can then be executed from command line. If the trace logger has not been configured already it is set to propagate=False so it needs to be activated separately.

Environment considerations

When using the client there are several limitations of your environment that could come into play.

When using an http load balancer you cannot use the Sniffing functionality - the cluster would supply the client with IP addresses to directly cnnect to the cluster, circumventing the load balancer. Depending on your configuration this might be something you don’t want or break completely.

In some environments (notably on Google App Engine) your http requests might be restricted so that GET requests won’t accept body. In that case use the send_get_body_as parameter of Transport to send all bodies via post:

from elasticsearch import Elasticsearch
es = Elasticsearch(send_get_body_as='POST')


Copyright 2013 Elasticsearch

Licensed under the Apache License, Version 2.0 (the “License”); you may not use this file except in compliance with the License. You may obtain a copy of the License at

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an “AS IS” BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

Indices and tables